getzep/graphiti

What is it?

Key features

• 01Tracks fact validity windows and invalidates outdated facts instead of deleting them
• 02Keeps provenance from every derived fact back to its source episode
• 03Supports incremental updates without full graph recomputation
• 04Hybrid retrieval over embeddings, BM25 keywords and graph traversal
• 05Custom entity and edge types defined with Pydantic models
• 06Pluggable backends including Neo4j, FalkorDB and Amazon Neptune

Use cases

• Giving a chatbot or agent long-term memory of user preferences that change over time
• Building a knowledge base that merges structured records and documents while keeping history
• Answering point-in-time questions such as what was true last quarter
• Prototyping temporal graph RAG with a local model through an OpenAI-compatible endpoint

Quick start

1. 1

   Install the library with pip

   pip install graphiti-core
2. 2

   Set an OPENAI_API_KEY environment variable (Graphiti defaults to OpenAI)

3. 3

   Start a graph database such as FalkorDB via Docker for the quickstart

   docker run -p 6379:6379 -p 3000:3000 -it --rm falkordb/falkordb:latest
4. 4

   Run the quickstart example from the examples/quickstart directory

Things to know

• !Requires a graph database such as Neo4j, FalkorDB, or Neptune, plus an LLM API key, so setup takes real effort.
• !Self-hosted only; there is no managed hosting from this project.
• !Works best with LLMs that support structured output; smaller models may cause ingestion failures.
• !Default concurrency is low to avoid rate limits, so ingestion can be slow until tuned.

Topics

FAQ

What is graphiti?

Python framework for building temporal knowledge graphs that track how facts change over time, used as memory and context for AI agents. Graphiti builds context graphs for AI agents from conversations, structured records and unstructured text. Each fact is stored as an entity-relationship-entity edge with a validity window, so when information changes the old fact is invalidated rather than deleted, and both current and historical states can be queried. Every derived fact keeps a link back to the raw episode that produced it.

Is graphiti free to use? What license does it have?

graphiti is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.

Is graphiti still maintained?

Actively maintained. The last commit to graphiti was on Oct 20, 2026.

What can I use graphiti for?

Giving a chatbot or agent long-term memory of user preferences that change over time Building a knowledge base that merges structured records and documents while keeping history Answering point-in-time questions such as what was true last quarter Prototyping temporal graph RAG with a local model through an OpenAI-compatible endpoint

How do I install graphiti?

Install the library with pip: pip install graphiti-core Set an OPENAI_API_KEY environment variable (Graphiti defaults to OpenAI) Start a graph database such as FalkorDB via Docker for the quickstart: docker run -p 6379:6379 -p 3000:3000 -it --rm falkordb/falkordb:latest Run the quickstart example from the examples/quickstart directory

Where can I get graphiti?

The source code and installation instructions are on GitHub: https://github.com/getzep/graphiti